A method for rapid prediction of 28-day compressive strength of cement

Abstract

The application belongs to the field of computing intelligence, and particularly relates to a method for rapidly predicting 28-day compressive strength of cement, which starts from information of cement composition components, simultaneously considers influence of lime saturation coefficient on the result, and corrects 28-day compressive strength of cement by using 3-day compressive strength of cement; a BP neural network and a GM(1, N) sub-prediction module are constructed, and an IMODE algorithm is used to optimize module weights, so as to strengthen generalization performance of the model; the method does not depend on requirements of the Netherlands weight maturity on a production environment, and considers unique production conditions and processing differences of different manufacturers, so as to adaptively correct prediction deviation. The application can provide a guiding role in the cement production process, and effectively improve prediction accuracy of 28-day compressive strength of cement and help shorten a detection period.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of computing intelligence, and particularly relates to a method for rapidly predicting 28-day compressive strength of cement, which is based on GM(1, N) optimized BP neural network. BACKGROUND

[0002] The cement compressive strength is one of important evaluation indexes of cement quality, but the detection cycle of mortar compressive strength needs 28 days to complete, and the long detection cycle has a malignant influence on the production of cement. In order to accelerate the construction period and improve the efficiency, many engineering and technical personnel have carried out a large amount of research on the method of cement compressive strength.

[0003] At present, there are many prediction methods of cement compressive strength, for example, Xing Feng et al. designed a 24-hour rapid prediction method of cement compressive strength, the compressive strength of different ages was tested within the first 48 hours of 55 DEG C humid heat curing, and the relationship between the weight maturity and the strength was established to predict the 28-day compressive strength of cement (Xing Feng, Luo Qiling, Han Ningxu. A method for rapidly predicting 28-day compressive strength of cement mortar [P]. Guangdong: CN101929930B, 2012-09-05.), and then the compressive strength was tested at the age of 9h and 24h of 55 DEG C humid heat curing, and the relationship between the weight maturity and the strength was established to predict the 28-day compressive strength of cement, so that the prediction cycle was shortened from 48 hours to 24 hours on the basis of ensuring the prediction accuracy (Xing Feng, Luo Qiling, Wang Weilun, Han Ningxu, Long Wujian. A 24-hour rapid prediction method of cement compressive strength [P]. Guangdong: CN103674705A, 2014-03-26.). However, the two methods mentioned above depend on the weight maturity of the Netherlands, also known as C value maturity, and need to refer to the value of foreign cement weight coefficient C value to simplify the strength detection process, but cannot consider the unique production conditions and processing differences of different manufacturers, and cannot adaptively correct the deviation between the standard prediction value and the true value in the application process.

[0004] In order to enable the prediction method to machine learning on the existing data of specific cement manufacturers, thereby improving the prediction accuracy, Wu Xianguo et al. applied the random forest method to the prediction of cement compressive strength (Wu Xianguo, Liu Pengcheng, Chen Hongyu, et al. High-performance concrete compressive strength prediction based on random forest [J]. Concrete, 2022(1): 17-20.), Zhang Dapeng et al. selected the multivariate regression equation method (Zhang Dapeng, Wang Junlong, Tao Tai, et al. Application of multivariate regression equation in 28d cement compressive strength prediction [J]. Cement Technology, 2021(3): 63-68.), Ma Qingxiu selected the support vector machine method (Ma Qingxiu. Cement compressive strength prediction based on support vector machine [J]. Highway and Automobile, 2009(3): 93-96.), Wang Jizong et al. selected the BP neural network method (Wang Jizong, Ni Hongguang. Cement compressive strength prediction based on BP neural network [J]. Journal of Silicate, 1999(4): 26-32.), these methods have made beneficial exploration for the improvement of the prediction accuracy of cement compressive strength, and in the application process of specific cement manufacturers, with the increase of the number of uses, the stored cement sample data is also more and more, every time the compressive strength of a new cement sample is predicted, the latest data can be used to construct a model instance for prediction, so as to adaptively correct the deviation between the standard prediction value and the true value. However, the generalization performance and prediction accuracy of a single simple model still have certain space for improvement. The IMODE algorithm is an improved multi-algorithm differential evolution algorithm based on the standard differential evolution model, which has excellent generalization performance (Sallam KM, Elsayed SM, Chakrabortty RK, et al. Improved multi-operator differential evolution algorithm for solving unconstrained problems [C] / / 2020 IEEE Congress on Evolutionary Computation (CEC). IEEE, 2020: 1-8.), which is applied to the weight optimization process of the sub-model, which helps to improve the generalization performance of the prediction method.

[0005] In summary, the rapid detection of 28-day cement compressive strength helps to improve the production efficiency of cement manufacturers, in order to adapt to the differences in production conditions of various manufacturers, it is necessary to use machine learning method to predict 28-day cement compressive strength. The prediction accuracy of 28-day cement compressive strength has important influence on construction safety, and the generalization performance and prediction accuracy of the prediction method need to be further improved.

[0006] Therefore, it is necessary to provide a GM(1, N) optimized BP neural network method for rapid prediction of 28-day cement compressive strength to solve the above technical problems. SUMMARY

[0007] In order to achieve the above-mentioned purpose, the application discloses a method for rapidly predicting 28-day compressive strength of cement, which is based on GM(1, N) optimized BP neural network, and specifically comprises the following steps:

[0008] S1: loading a compressive strength database required for constructing a model for predicting 28-day compressive strength of cement, wherein each cement sample in the database is a row of data, and the data entries include multiple main cement component entries, lime saturation coefficient, 3-day compressive strength of cement, and 28-day compressive strength of cement;

[0009] S2: constructing a BP neural network model, which is a compressive strength prediction module, taking the multiple main cement component data, lime saturation coefficient, and 3-day compressive strength of cement as inputs, taking 28-day compressive strength of cement as output, using a trainbfg training algorithm to train the model, and finally calculating the mean square error of the BP neural network model;

[0010] S3: constructing a GM(1, N) model, which is a compressive strength prediction module, using the main cement component data, lime saturation coefficient, 3-day compressive strength of cement, and 28-day compressive strength of cement data to construct the GM(1, N) model, performing level ratio test, establishing a time series of 28-day compressive strength data, calculating the level ratio and performing level ratio judgment, subsequently constructing a data matrix and a data vector, and finally calculating the correlation degree of the main cement components, lime saturation coefficient, and 3-day compressive strength of cement, and finally calculating the mean square error of the GM(1, N) model;

[0011] S4: constructing a GM(1, N) optimized BP neural network compressive strength prediction model, using an IMODE algorithm to optimize the weights of the BP neural network model and the GM(1, N) model in the GM(1, N) optimized BP neural network compressive strength prediction model, and after optimization, using the optimized model to calculate the predicted value of 28-day compressive strength of cement of a to-be-predicted cement sample;

[0012] S5: using existing equipment of the "Cement Strength Rapid Test Method" JC / T738-2004, detecting the true value of 28-day compressive strength of cement of the to-be-predicted cement sample according to the strength test method of GB / T17671-1999 "Cement Mortar Strength Test Method (ISO)", and storing the main cement component data, lime saturation coefficient, 3-day compressive strength of cement, and 28-day compressive strength of cement of the cement sample into a historical operation database, wherein the main cement component entries include the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample.

[0013] In order to effectively utilize historical data to improve prediction accuracy, in step S1, the compressive strength database required for constructing the prediction model of the 28-day compressive strength of the cement is loaded according to the following steps:

[0014] Step 1-1: Load the compressive strength database from the historical operation database. If there is no historical operation database, construct an empty table of the compressive strength database. Each cement sample in the compressive strength database is a row of data, and the data entries include multiple cement main component entries, a lime saturation coefficient, a 3-day compressive strength of the cement, and a 28-day compressive strength of the cement. The loaded multiple cement main component entries include SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF. The value corresponding to each cement main component entry represents the mass percentage of the component;

[0015] Step 1-2: Run the method with data of at least 30 cement samples. Check the number of data rows in the compressive strength database. If the number is less than 30, go to step 1-3. Otherwise, go to step 1-7.

[0016] Step 1-3: Use the cement main components to construct a kind of cement sample whose main component values are not completely the same as those of any cement sample in the compressive strength database. Insert it into the first row of the compressive strength database. At this time, the other existing data automatically moves down one row. In the multiple cement main component entries in the first row of the compressive strength database, store the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample. Perform normalization operation on the data of the multiple cement main component entries. Multiply the result of the normalization operation by 100 so that the multiple cement main component entries are all positive and the sum is 100.

[0017] Step 1-4: Calculate the lime saturation coefficient of this cement sample, which is used to represent the degree to which silica in clinker is saturated with calcium oxide to form tricalcium silicate. Calculate the lime saturation coefficient according to the following formula:

[0018] KH = (CaO - 1.65 Al2O3 - 0.35 Fe2O3) / 2.8 SiO2

[0019] Where KH represents the lime saturation coefficient, and CaO, Al2O3, Fe2O3, and SiO2 are the mass percentages of the corresponding oxides in the clinker.

[0020] Step 1-5: using the existing equipment of "Cement Strength Rapid Inspection Method" JC / T738-2004, according to the strength test method of GB / T17671-1999 "Cement Mortar Strength Test Method (ISO)", the current cement sample is detected for cement 3-day compressive strength, cement 28-day compressive strength;

[0021] Step 1-6: the lime saturation factor, cement 3-day compressive strength, and cement 28-day compressive strength of the current cement sample are stored in the corresponding data entries of the first row of the compressive strength database. Check the number of data rows in the compressive strength database. If it is less than 30, go to step 1-3, otherwise go to step 1-7;

[0022] Step 1-7: insert a blank row in the first row of the compressive strength database. At this time, the other existing data automatically moves down one row. The composition information of the cement sample whose cement 28-day compressive strength value needs to be predicted is stored in the corresponding position of the first row of the compressive strength database. The lime saturation factor of the cement sample is calculated according to the method of step 1-4, and the cement 3-day compressive strength of the cement sample is determined according to the method of step 1-5. The lime saturation factor and the cement 3-day compressive strength value are stored in the corresponding position of the first row of the compressive strength database, and the cement 28-day compressive strength value is empty.

[0023] In order to preliminarily mine the cement sample data, in the step S2, a BP neural network model is constructed according to the following steps:

[0024] Step 2-1: a BP neural network model is constructed, which has a specific structure of a single hidden layer BP neural network model containing 10 neurons, and uses a sigmoid activation function;

[0025] Step 2-2: check the compressive strength database, and only select the data rows without empty values for the construction of the BP neural network model. All available data rows are randomly divided into two sets according to the proportion of 80% and 20%. The set with a proportion of 80% is called the training set, and the set with a proportion of 20% is called the test set;

[0026] Step 2-3: Each data row includes a plurality of cement main component items, a lime saturation factor, a cement 3-day compressive strength, and a cement 28-day compressive strength, wherein the cement main component items include mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample. An input layer and an output layer are constructed for the BP neural network model, wherein the input layer is a single-layer structure including 12 input nodes, and in the 12 input nodes, the first 10 input nodes correspond to the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF, respectively, and the 11th and 12th input nodes correspond to the lime saturation factor and the cement 3-day compressive strength, respectively. The output layer of the BP neural network model is a single-layer structure including one output node, wherein the first output node corresponds to the cement 28-day compressive strength.

[0027] Step 2-4: The constructed BP neural network model is trained using the training set, and the model is tested using the test set. During the training process, the trainbfg training algorithm is used to train the model. After the training is completed, the mean square error of the BP neural network model is calculated.

[0028] In order to conduct in-depth data mining on the cement sample data, in step S4, a compressive strength prediction model of the GM(1, N) optimized BP neural network is constructed according to the following steps:

[0029] Step 4-1: The predicted values of the cement 28-day compressive strength of each cement sample in the compressive strength database are derived from the two compressive strength prediction modules of the BP neural network model and the GM(1, N) model constructed in steps S2 and S3.

[0030] Step 4-2: An optimization model based on the IMODE algorithm is constructed, wherein the target function to be optimized is:

[0031]

[0032] wherein G is the target function to be optimized, l is the total number of cement samples in the compressive strength database that do not have empty data values, S i represents the i-th cement sample, g(S i ) represents the square of the error value between the predicted value and the true value of the i-th cement sample, x i and y i represent the true value and the predicted value of the cement 28-day compressive strength of the i-th cement sample, respectively;

[0033] Step 4-3: the parameters of the IMODE algorithm are set as: the minimum population minN = 4, the ratio of the external archive to the population size aRate = 2.6. In the optimization process, the IMODE algorithm is initialized according to the following parameters: the population size N = 50, the maximum number of evaluations maxFE = 1000. The IMODE algorithm is run in the direction of minimizing the objective function G. In the initial state, the evaluation number FE = 0, and the evaluation number FE is increased by one each time the value of the objective function G is calculated. When FE is greater than or equal to maxFE, the IMODE algorithm ends running;

[0034] Step 4-4: the multiple cement main component items of the cement sample to be predicted, the lime saturation coefficient and the cement 3-day compressive strength are input into the compressive strength prediction model of the GM(1, N) optimized BP neural network obtained after the IMODE algorithm, wherein the cement main component items include the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A and C4AF of the cement sample. The predicted value of the cement 28-day compressive strength of the cement sample to be predicted is calculated according to the following formula:

[0035]

[0036] Wherein, y is the predicted value of the cement 28-day compressive strength of the cement sample to be predicted, y BP , y GM are the predicted values of the cement 28-day compressive strength of the cement sample to be predicted obtained by the two prediction modules of the BP neural network model and the GM(1, N) model respectively, ω BP , ω GM are the weight values of the BP neural network model and the GM(1, N) model obtained by the IMODE algorithm.

[0037] The beneficial effects of the present application are: compared with the prior art, the rapid prediction method of the cement 28-day compressive strength provided by the present application starts from the cement component information, considers the influence of the lime saturation coefficient on the result, and corrects the cement 28-day compressive strength by the cement 3-day compressive strength; the BP neural network and the GM(1, N) sub-prediction modules are constructed, the IMODE algorithm is used for module weight optimization, and the generalization performance of the model is strengthened; the method does not depend on the mature degree of the Dutch weight, considers the unique production conditions and processing differences of different manufacturers, and thus the prediction deviation is corrected adaptively.

[0038] The present application can provide guidance in the cement production process and effectively improve the prediction accuracy of the cement 28-day compressive strength and help shorten the detection period. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 Method flow chart for one embodiment of the present application. DETAILED DESCRIPTION

[0040] In order to deepen the understanding of the present application, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments, which are only used to explain the present application and do not constitute a limitation on the protection scope of the present application.

[0041] As shown in Figure 1 A method for rapidly predicting the 28-day compressive strength of cement, the method comprising:

[0042] S1: load the compressive strength database required for building the model for predicting the 28-day compressive strength of cement, each cement sample in the database being a row of data, the data entries of which include a plurality of cement main component entries, a lime saturation coefficient, a 3-day compressive strength of cement, and a 28-day compressive strength of cement, wherein the compressive strength database required for building the model for predicting the 28-day compressive strength of cement is loaded according to the following steps:

[0043] Step 1-1: load the compressive strength database from the historical operation database, and if there is no historical operation database, build an empty table of the compressive strength database. Each cement sample in the compressive strength database is a row of data, the data entries of which include a plurality of cement main component entries, a lime saturation coefficient, a 3-day compressive strength of cement, and a 28-day compressive strength of cement, wherein the loaded plurality of cement main component entries include SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF, and the value corresponding to each cement main component entry represents the mass percentage of the component;

[0044] Step 1-2: run the data of at least 30 cement samples required by the present method, check the number of data rows in the compressive strength database, and when it is less than 30, go to step 1-3, otherwise go to step 1-7;

[0045] Step 1-3: using the cement main components, build a kind of cement sample whose main component entries are not completely the same as those of any cement sample in the compressive strength database, and insert it into the first row of the compressive strength database, at this time, the other existing data automatically moves down one row. In the plurality of cement main component entries in the first row of the compressive strength database, store the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample, and perform a normalization operation on the data of the plurality of cement main component entries, multiply the result of the normalization operation by 100, so that the plurality of cement main component entries are all positive values and the sum is 100;

[0046] Step 1-4: Calculate the lime saturation factor of this cement sample, which is used to represent the degree of saturation of silica in the clinker by calcium oxide into tricalcium silicate, according to the following formula to calculate the lime saturation factor:

[0047] KH = (CaO - 1.65Al203- 0.35Fe203) / 2.8Si02

[0048] Wherein, KH represents the lime saturation factor, CaO, Al203, Fe203, Si02 are the mass percentage of the corresponding oxide in the clinker;

[0049] Step 1-5: Use the existing equipment of "Cement Strength Rapid Test Method" JC / T738-2004, according to the strength test method of "Cement Mortar Strength Test Method (ISO)" GB / T17671-1999, to detect the 3-day compressive strength of the current cement sample, the 28-day compressive strength of the cement;

[0050] Step 1-6: Store the lime saturation factor, 3-day compressive strength of the current cement sample, 28-day compressive strength of the cement into the corresponding data entries of the first row of the compressive strength database, check the number of data rows in the compressive strength database, when it is less than 30, go to step 1-3, otherwise go to step 1-7;

[0051] Step 1-7: Insert a blank row in the first row of the compressive strength database, at this time, other existing data automatically moves down one row. Store the ingredient information of the cement sample whose 28-day compressive strength value needs to be predicted in the multiple cement main component data of the first row of the compressive strength database. Calculate the lime saturation factor for the cement sample according to the method of step 1-4, and determine the 3-day compressive strength of the cement sample according to the method of step 1-5. Store the values of the lime saturation factor and the 3-day compressive strength of the cement into the corresponding positions of the first row of the compressive strength database, and leave the value of the 28-day compressive strength of the cement empty;

[0052] S2: Build a BP neural network model for this compressive strength prediction module, use multiple cement main component data, lime saturation factor, and 3-day compressive strength of cement as input, and use 28-day compressive strength of cement as output. Use the trainbfg training algorithm for model training, and finally calculate the mean square error of the BP neural network model. Wherein, according to the following steps to build a BP neural network model for this compressive strength prediction module:

[0053] Step 2-1: Build a BP neural network model, which specifically includes a single hidden layer BP neural network model containing 10 neurons, using a sigmoid activation function;

[0054] Step 2-2: Check the compressive strength database, only select the data rows without null values for the construction of the BP neural network model, and randomly divide all available data rows into two sets in the ratio of 80% and 20%, with the set with a proportion of 80% being referred to as the training set and the set with a proportion of 20% being referred to as the test set;

[0055] Step 2-3: Each data row includes multiple cement main component items, lime saturation coefficient, cement 3-day compressive strength, and cement 28-day compressive strength, wherein the cement main component items include the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample. The input layer and the output layer are constructed for the BP neural network model, wherein the input layer is a single-layer structure including 12 input nodes, among which the first 10 input nodes correspond to the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF, and the 11th and 12th input nodes correspond to the lime saturation coefficient and the cement 3-day compressive strength, respectively. The output layer of the BP neural network model is a single-layer structure including 1 output node, wherein the first output node corresponds to the cement 28-day compressive strength;

[0056] Step 2-4: The constructed BP neural network model is trained using the training set and tested using the test set. During the training process, the trainbfg training algorithm is used for model training. After training is completed, the mean square error of the BP neural network model is calculated;

[0057] S3: Construct the GM(1, N) model, which is a compressive strength prediction module, using cement main component data, lime saturation coefficient, cement 3-day compressive strength, and cement 28-day compressive strength data to construct the GM(1, N) model, perform level ratio testing, establish a 28-day compressive strength data time series, calculate the level ratio and perform level ratio judgment, then construct a data matrix and a data vector, and finally calculate the correlation degree of each cement main component, lime saturation coefficient, and cement 3-day compressive strength. Finally, the mean square error of the GM(1, N) model is calculated;

[0058] S4: Construct the GM(1, N) optimized BP neural network compressive strength prediction model, and use the IMODE algorithm to optimize the weights of the BP neural network model and the GM(1, N) model in the GM(1, N) optimized BP neural network compressive strength prediction model. After optimization, the predicted value of the cement 28-day compressive strength of the cement sample to be predicted is calculated using the optimized model. The GM(1, N) optimized BP neural network compressive strength prediction model is constructed according to the following steps:

[0059] Step 4-1: Derive the prediction values of the 28-day compressive strength of each cement sample in the compressive strength database from the BP neural network model and the GM(1, N) model constructed in steps S2 and S3;

[0060] Step 4-2: Construct an optimization model based on the IMODE algorithm, where the target function to be optimized is:

[0061]

[0062] where G is the target function to be optimized, I is the total number of cement samples in the compressive strength database without empty data values, S i represents the i-th cement sample, g(S i ) represents the square of the error value between the predicted value and the true value of the i-th cement sample, x i and y i represent the true value and the predicted value of the 28-day compressive strength of the i-th cement sample, respectively;

[0063] Step 4-3: Set the parameters of the IMODE algorithm as follows: minimum population minN = 4, ratio of external archive to population size aRate = 2.6. During optimization, initialize the IMODE algorithm with the following parameters: population size N = 50, maximum evaluation times maxFE = 1000. Run the IMODE algorithm in the direction of minimizing the target function G. In the initial state, the evaluation times FE = 0, and each time the value of the target function G is calculated, the evaluation times FE is incremented by one. When FE is greater than or equal to maxFE, the IMODE algorithm stops running;

[0064] Step 4-4: Input the multiple cement main component items, lime saturation coefficient, and 3-day compressive strength of the cement sample to be predicted into the compressive strength prediction model of the GM(1, N) optimized BP neural network, where the cement main component items include the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample. Calculate the predicted value of the 28-day compressive strength of the cement sample to be predicted according to the following formula:

[0065]

[0066] where y is the predicted value of the 28-day compressive strength of the cement sample to be predicted, y BP and y GM are the predicted values of the 28-day compressive strength of the cement sample to be predicted obtained by the BP neural network model and the GM(1, N) model, respectively, ω BP and ωGM are weight values of the BP neural network model and the GM(1, N) model optimized by the IMODE algorithm respectively.

[0067] S5: using the existing equipment of "Cement Strength Rapid Test Method" JC / T738-2004, detecting the real value of the 28-day compressive strength of the cement sample to be predicted according to the strength test method of "Cement Mortar Strength Test Method (ISO)" GB / T17671-1999, and storing the cement main component data, lime saturation coefficient, 3-day compressive strength of cement, and 28-day compressive strength information of the cement sample into the historical operation database, wherein the cement main component entries include the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample.

[0068] The application will be further described below in combination with specific examples.

[0069] Example 1: Cement sample information and 28-day compressive strength prediction value thereof

[0070] Table 1 shows the composition information of a certain cement sample and the lime saturation coefficient KH value and 3-day compressive strength of the cement. The error rates of the predicted strength and the real strength of the 28-day compressive strength of the cement obtained by the BP neural network method, the GM(1, N) method, and the method are 0.98%, -0.36%, and -0.19% respectively.

[0071] Table 1 Cement sample information and 28-day compressive strength prediction value thereof

[0072]

[0073] Example 2: Cement sample information and 28-day compressive strength prediction value thereof

[0074] Table 2 shows the composition information of a certain cement sample and the lime saturation coefficient KH value and 3-day compressive strength of the cement. The error rates of the predicted strength and the real strength of the 28-day compressive strength of the cement obtained by the BP neural network method, the GM(1, N) method, and the method are 0.38%, -0.16%, and -0.08% respectively.

[0075] Table 2 Cement sample information and 28-day compressive strength prediction value thereof

[0076]

[0077] In the description of the application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application.

[0078] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A rapid prediction method for the 28-day compressive strength of cement, characterized in that, Includes the following steps: S1: Load the compressive strength database required to build a model for predicting the 28-day compressive strength of cement. Each cement sample in the database is a row of data, and its data entries include multiple cement main component entries, lime saturation coefficient, cement 3-day compressive strength, and cement 28-day compressive strength. S2: Construct a BP neural network model as the compressive strength prediction module. Take multiple main cement component data, lime saturation coefficient, and cement 3-day compressive strength as inputs and cement 28-day compressive strength as output. Use the trainbfg training algorithm to train the model and finally calculate the mean square error of the BP neural network model. S3: Construct the GM(1,N) model for compressive strength prediction. This module uses data on the main components of cement, lime saturation coefficient, 3-day compressive strength of cement, and 28-day compressive strength of cement to construct the GM(1,N) model. It performs a grade ratio test, establishes a time series of 28-day compressive strength data, calculates the grade ratio, and makes a grade ratio judgment. Then, it constructs a data matrix and data vector. Finally, it calculates the correlation between the main components of cement, lime saturation coefficient, and 3-day compressive strength of cement, and finally calculates the mean square error of the GM(1,N) model. S4: Construct a compressive strength prediction model using a GM(1,N) optimized BP neural network. Use the IMODE algorithm to optimize the weights of the two compressive strength prediction modules, the BP neural network model and the GM(1,N) model, in the GM(1,N) optimized BP neural network compressive strength prediction model. After optimization, use the optimized model to calculate the predicted value of the 28-day compressive strength of the cement sample to be predicted. S5: Detect the true value of the 28-day compressive strength of the cement sample to be predicted, and store the main cement component data, lime saturation coefficient, 3-day compressive strength, and 28-day compressive strength information of the cement sample into the historical operation database. The main cement component entries include the mass percentage of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample.

2. The rapid prediction method for the 28-day compressive strength of cement according to claim 1, characterized in that, In step S1, the compressive strength database required for constructing the model for predicting the 28-day compressive strength of cement is loaded according to the following steps: Step 1-1: Load the compressive strength database from the historical operation database. If there is no historical operation database, construct an empty table for the compressive strength database. Each cement sample in the compressive strength database is a row of data. Its data entries include multiple cement main component entries, lime saturation coefficient, cement 3-day compressive strength, and cement 28-day compressive strength. Among them, the loaded multiple cement main component entries include SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF. The value corresponding to each cement main component entry represents the mass percentage of that component. Step 1-2: Check the number of data rows in the compressive strength database. If it is less than 30, proceed to step 1-3; otherwise, proceed to step 1-7. Steps 1-3: Using the main components of cement, construct a cement sample whose value is not exactly the same as the main component entry of any cement sample in the compressive strength database. Insert it into the first row of the compressive strength database. At this time, other existing data will automatically move down one row. Store the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of this cement sample in multiple main component entries in the first row of the compressive strength database. Perform a normalization operation on the data of multiple main component entries, and multiply the result of the normalization operation by 100 so that all the main component entries are positive and the sum is 100. Steps 1-4: Calculate the lime saturation coefficient of this cement sample, which represents the degree to which silica in the clinker is saturated with calcium oxide to form tricalcium silicate; Steps 1-5: Using the existing equipment of "Rapid Test Method for Cement Strength" JC / T738-2004, and according to the strength test method of GB / T17671-1999 "Test Method for Cement Mortar Strength (ISO)", test the 3-day compressive strength and 28-day compressive strength of the current cement sample. Steps 1-6: Store the lime saturation coefficient, 3-day compressive strength, and 28-day compressive strength of the current cement sample into the corresponding data entries in the first row of the compressive strength database. Check the number of data rows in the compressive strength database. If it is less than 30, proceed to step 1-3; otherwise, proceed to step 1-7. Steps 1-7: Insert a blank row in the first row of the compressive strength database. At this time, other existing data will automatically move down one row. Store the composition information of the cement sample for which the 28-day compressive strength value of cement needs to be predicted. Store the data of multiple main cement component entries in the corresponding positions of the first row of the compressive strength database. Calculate the lime saturation coefficient for the cement sample according to the method in Steps 1-4. Measure the 3-day compressive strength of the cement sample according to the method in Steps 1-5. Store the lime saturation coefficient and the 3-day compressive strength value of cement in the corresponding positions of the first row of the compressive strength database. Leave the 28-day compressive strength value of cement blank.

3. The rapid prediction method for the 28-day compressive strength of cement according to claim 2, characterized in that, In steps 1-4, the lime saturation coefficient is calculated according to the following formula: KH=(CaO-1.65Al2O3-0.35Fe2O3) / 2.8SiO2 Wherein, KH represents the lime saturation coefficient, and CaO, Al2O3, Fe2O3, and SiO2 are the mass percentages of the corresponding oxides in the clinker.

4. The rapid prediction method for the 28-day compressive strength of cement according to claim 1, characterized in that, In step S2, a BP neural network model group is constructed as a compressive strength prediction module according to the following steps: Step 2-1: Construct a BP neural network model, specifically a single hidden layer BP neural network model containing 10 neurons, using the sigmoid activation function; Step 2-2: Check the compressive strength database and select only the data rows without null values ​​for the construction of the BP neural network model. Randomly divide all available data rows into two sets according to the ratio of 80% and 20%. The set with the 80% ratio is called the training set, and the set with the 20% ratio is called the test set. Steps 2-3: Each data row includes multiple entries for the main components of cement, lime saturation coefficient, 3-day compressive strength of cement, and 28-day compressive strength of cement. The entries for the main components of cement include the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample. This forms the input and output layers for the BP neural network model. The input layer is a single-layer structure with 12 input nodes. The first 10 input nodes correspond one-to-one with the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF. The 11th and 12th input nodes correspond one-to-one with the lime saturation coefficient and the 3-day compressive strength of cement, respectively. The output layer of the BP neural network model is a single-layer structure with one output node, which corresponds to the 28-day compressive strength of cement. Steps 2-4: Train the constructed BP neural network model using the training set and validate the model using the validation set; during training, use the trainbfg training algorithm to train the model; after training, calculate the mean square error of the BP neural network model.

5. The rapid prediction method for the 28-day compressive strength of cement according to claim 1, characterized in that, In step S4, a GM(1,N) optimized BP neural network model for predicting compressive strength is constructed according to the following steps: Step 4-1: Export the BP neural network model and GM(1,N) model constructed in steps S2 and S3 as the two compressive strength prediction modules to predict the 28-day compressive strength of cement samples in the compressive strength database. Step 4-2: Construct an optimization model based on the IMODE algorithm; Step 4-3: Optimize the IMODE algorithm; The parameters of the IMODE algorithm are set as follows: minimum population value minN = 4, ratio of external archive to population size aRate = 2.

6. During the optimization process, the IMODE algorithm is initialized with the following parameters: population size N = 50, maximum number of evaluations maxFE = 1000. The optimization direction is to minimize the objective function G. The IMODE algorithm is run. In the initial state, the number of evaluations FE = 0. Each time the value of the objective function G is calculated, the number of evaluations FE is incremented by one. When FE is greater than or equal to maxFE, the IMODE algorithm ends. Step 4-4: Input the main cement components, lime saturation coefficient, and 3-day compressive strength of the cement sample to be predicted into the GM(1,N) optimized BP neural network compressive strength prediction model obtained by the IMODE algorithm. The main cement components include the mass percentages of SiO2, Al2O3, Fe2O3, CaO, MgO, SO3, C3S, C2S, C3A, and C4AF of the cement sample.

6. The rapid prediction method for the 28-day compressive strength of cement according to claim 5, characterized in that, In step S4, step 4-2: Construct an optimization model based on the IMODE algorithm, wherein the objective function to be optimized is: Where G is the objective function to be optimized, l is the total number of cement samples in the compressive strength database that do not contain null data values, and S i Let g(S) represent the i-th cement sample. i x represents the square of the error between the predicted value and the actual value of the i-th cement sample. i and y i Let represent the actual value and the predicted value of the 28-day compressive strength of the i-th cement sample, respectively.

7. The rapid prediction method for the 28-day compressive strength of cement according to claim 5, characterized in that, In step 4-3, the IMODE algorithm is optimized: the parameters of the IMODE algorithm are set to the minimum population value minN = 4 and the ratio of external archive to population size aRate = 2.

6. During the optimization process, the IMODE algorithm is initialized with the following parameters: population size N = 50, maximum number of evaluations maxFE = 1000. The optimization direction is to minimize the objective function G. The IMODE algorithm is run. In the initial state, the number of evaluations FE = 0. Each time the value of the objective function G is calculated, the number of evaluations FE is incremented by one. When FE is greater than or equal to maxFE, the IMODE algorithm ends.

8. The rapid prediction method for the 28-day compressive strength of cement according to claim 5, characterized in that, In step 4-4, the predicted value of the 28-day compressive strength of the cement sample to be predicted is calculated according to the following formula: Where y is the predicted value of the 28-day compressive strength of the cement sample to be predicted, yBP and yGM are the predicted values ​​of the 28-day compressive strength of the cement sample to be predicted obtained by the BP neural network model and the GM(1,N) model, respectively, and ω BP ω GM These are the weight values ​​of the BP neural network model and the GM(1,N) model obtained by optimization using the IMODE algorithm, respectively.

Images